Will the human mind remain unique in the age of AI? | | Books

Uuntil recently, we humans have been overreacting to our abilities. No other animals play board games, write notes or do math proofs. But recently, advances in AI seem to challenge our self-esteem as the smartest things around. AI systems not only beat us at the hardest games, but they can also write polished prose and win medals in math. Tech giants promise us that super-powerful AI is just around the corner. So, in the age of AI, is the human mind still unique, or is it just running?

Talking about superhuman AI assumes that intelligence is one scale. My parents used to put a sign above me and my younger brother on the door of our wardrobe. Every year he got a little closer to me, until one improbable year happened and he surpassed me (he’s now 6ft 3in). The current time feels like that, as we look at our new brothers and sisters and worry that they might find us.

But wisdom is not like height. There is only one way to be tall, but there are many ways to be smart. Just looking at other animals tells us about it. Even though humans are great, we are still fascinated by how birds fly, how ants work together, and how spiders behave. Each of these animals is designed by its environment to be intelligent in some way.

People are not different. Our thoughts are shaped by our biology. We only live for a few decades and we have to learn all we can learn and do all we can do in a short time. Learning and doing will all happen directly in the kilo or neurons that are locked inside our skulls bones. We can simply share our thoughts with others by making noises with our mouths or tapping and shaking our fingers.

AI machines do not face these problems. He can process more than any human can see in a lifetime. They can increase their power by using more computers. And they can easily share what they see and learn with other machines.

Our short lives, squishy brains and loud mouths may seem inferior compared to machines: in fact, these are the things that make us unique, and we will continue to do so.

That’s human wisdom the answer to our failures. To get the most out of our lives, we have an amazing ability to learn from small experiences. Yes, AlphaGo can beat the best players, but it was trained on a lot of human games. Yes, ChatGPT can communicate well, but it is based on thousands of years of language. No AI system can make decisions with the skills of a five-year-old human when faced with massive amounts of data.

This also applies to our limited brain and our communication skills. We can’t just spin up another computer when we need extra processing power. This means that we must be good at recognizing patterns in work and use our attention wisely. Relying on word of mouth is difficult. To deal with this, we have developed tools – language, writing, teaching, and science – to integrate knowledge between people and time. This means we need to be mindful of what’s going on in other people’s heads and work together to achieve our shared goals.

Because humans and machines face different challenges, we should expect them to find different ways to solve the problems they face. Although modern AI systems are beginning to do many things that humans can do, they often do them differently. The answers they get are shaped by their own experiences and tools.

Here is a simple example. How many letters are there in this sequence: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa? For a person, it is not difficult to answer – you can just count. For an AI system, it’s difficult. They are constrained by the way they represent language and the way they teach it. They tend to divide words into parts (called “signs”), which can prevent them from answering spelling questions. And they tend to target tokens that appear frequently in their notifications as responses. Us found OpenAI’s version of GPT-4, which was hailed as showing “traces of common intelligence”, he was ready to answer the question correctly when given 30 letters instead of 29. Why?

This is not the only area where AI faces challenges. Imagine you are helping a pharmacist. They require chemicals with 785 parts per million (ppm). Two test tubes are available: one has 685 ppm and the other 791 ppm. Your task is to determine which tube provides the correct concentration and dose you need. We hope you choose 791 ppm. However, sometimes even the AI ​​guidance system selects 685 ppm. Why? Because the artificial neural networks used in AI systems tend to lump things together. When there are two answers, they choose something in between. The number 785 can be represented as a series of numbers (“7”, “8”, and “5”) or as a number (seven hundred and eighty five). If it’s a string, 785 is very similar to 685 – they’re just single digits. But if it is a quantity, then it is very similar to 791. Mixing the two answers can have a big effect.

Human intelligence is based on a series of continuous events that are used to train AI systems. We use our brains to put diapers on babies, play chess, prove theorems, cook dinner, write novels and make monkeys. AI machines are usually trained to do only one thing – you can ask ChatGPT for advice on diapers, but they can’t pick up a squirming baby. The human brain and capable of all this because they evolved into a world that gives us all these problems, leaving us ready to learn the things we can hope to do in one person’s life.

Our limited lives, limited brains and limited ability to communicate have shaped the nature of human intelligence. So we can expect that human thinking will continue to be somewhat unique, even as we continue to develop intelligent machines. Remember: intelligence isn’t just one thing, it’s AI that reaches a point where people have left it at home.

This line of thinking should make us question claims of superhuman AI. Paying attention to the differences in constraints, training and hardware also proves that: AI will not be better than humans at everything. Rather, it will be better than humans in some ways and worse in others. AI and human mind will just be different from each other. And as brothers and sisters, maybe we can learn to treat each other not as competitors, but as friends.

Tom Griffiths is the professor of more technology at Princeton University and the author of Laws of Mind (William Collins)

Another reading

The World Appears by Michael Pollan (Allen Lane, £25)

If Everyone Builds, Everyone Dies by Eliezer Yudkowsky (Bodley Head, £22)

Being You: A New Science of Consciousness by Anil Seth (Faber, £12.99)